Detection of Defective and Non-defective Fuse Configurations of the Fuse Boxes used in Wiring Harnesses in Automobiles with Deep Learning

The fast and accurate automated quality visual inspection is increasingly gaining importance in manufacturing and product quality control for product efficiency. To effectively detect faults or defects in products, main methods focus on handcrafted optical features. An important part in all automobiles is the fuse box which provides protection to all electronic devices in a vehicle such as headlights, AC, radio, indicators, etc. The fuse box contains on several fuses placed in their position with the help of color coding and ampere rating. The current practice is to conduct manual inspection, which leads to high production cost and other quality issues. Our approach focuses on use of a Convolutional Neural Network to extract powerful features with less prior knowledge about the images for defect detection. This can improve the quality of the fuse boxes shipped for final assembly of the vehicle and reduce its cost of manufacturing and testing. In this report, we suggest a method for Implementation of Convolutional Neural Network to classify the fuse box configurations into correct and wrong classes based on color coding of fuses which can be extended in a similar way to other manufacturing quality control problems.